AI-based financial core system network security internal threat detection method and system
By adopting AI-based risk information and behavior detection modules in the financial core system and combining automated alarm tools, internal threat detection problems in the financial core system are solved, threat identification and disposal efficiency is improved, and operation and maintenance costs are reduced.
Patent Information
- Application Number
- CN202410936558.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-12
AI Technical Summary
It is difficult for the existing technology to effectively detect and respond to internal threats in the financial core system, and conventional security equipment is difficult to identify and deal with internal attacks caused by internal vulnerabilities, identity impersonation and illegal power raising.
Using AI-based risk information, asset linkage and behavior detection modules, combined with automated alarm script tools, the training risk information key information extraction model and internal abnormal access behavior detection model can detect and judge attack behavior in real time, and generate risk information analysis reports , automatically obtain critical risk vulnerabilities and block abnormal behavior.
It improves the accuracy and efficiency of identification of internal threats in network security in the financial core system, saves the execution costs of security operation and maintenance personnel, and realizes automatic detection and handling of internal attack behaviors.
Smart Images

Figure CN118944916A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of AI and network security technology, and in particular, to an AI-based financial core system network security internal threat detection method and system. Background Art
[0002] The current external threat detection methods based on feature libraries and attack detection are relatively mature, but in the scenario of financial core systems, conventional security protection equipment (IPS / IDS / WAF / NGFW) and situational awareness are difficult to deal with internal security threats caused by internal vulnerabilities and internal behaviors. Internal attacks caused by untimely repair of internal vulnerabilities, identity fraud, and illegal privilege escalation are endless. In order to comprehensively target the pain points of internal threat detection in the network security of financial core systems, through AI-based risk information, asset linkage, behavior detection and other modules, combined with automated alarm script tools for real-time reporting, timely and automatic acquisition of key risk vulnerabilities, self-checking of internal components, and blocking of internal abnormal behaviors, the accuracy of identifying and handling internal threats to the network security of financial core systems can be improved, and the execution costs of security operation and maintenance personnel and related IT personnel can be saved. Summary of the invention
[0003] In view of the defects in the prior art, the purpose of the present invention is to provide an AI-based financial core system network security internal threat detection method and system.
[0004] According to the present invention, a method for detecting internal threats to network security of a financial core system based on AI includes:
[0005] Step S1: Real-time risk information data uses the trained risk information key information extraction model to extract key information that meets preset requirements;
[0006] Step S2: The information reporting model generates a risk information analysis report based on the extracted key information that meets the preset requirements;
[0007] Step S3: Real-time full traffic information uses the trained internal abnormal access behavior detection model to detect and judge attack behaviors;
[0008] The risk information key information extraction model is an AI training model built based on the AI integration framework of langchain, the ERNIE-Bot 4.0 large model, and the web framework, with key fields that meet the preset requirements;
[0009] The information reporting model outputs a risk information report based on key information fields through an AI model;
[0010] The internal abnormal access behavior detection model is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network built through a PyTorch framework.
[0011] Preferably, the step S1 adopts:
[0012] Step S1.1: Obtain historical risk information, and preprocess the acquired historical risk information to obtain preprocessed historical risk information;
[0013] Step S1.2: Divide the preprocessed historical information into a training set and a test set;
[0014] Step S1.3: Use the training set to train the risk information key information extraction model to obtain the trained risk information key information extraction model; use the test set to verify and optimize the parameters of the trained risk key information extraction model;
[0015] Step S1.4: Use the test set data and real-time information data to confirm that the trained risk information key information extraction model meets the risk information key information extraction expectations;
[0016] Step S1.5: Utilize the trained risk information key information extraction model to extract key field information that meets preset requirements based on the acquired real-time information data.
[0017] Preferably, step S1.1 includes: performing cleaning and conversion processing on the information data of the official website announcement of security vendors that meet preset requirements based on NIST-NVD, CNVD, and obtain processed historical risk information data.
[0018] Preferably, the step S1.3 adopts:
[0019] Step S1.3.1: compile a document parsing model using components including Pandas, libreoffice, win32com, and PyMuPDF to extract text information from input documents in word, excel, and PDF formats;
[0020] Step S1.3.2: compile an image recognition model using components including paddlepaddle, opencv, and paddleocr to perform image recognition and text information extraction on input images in jpg format;
[0021] Step S1.3.3: Build front-end and back-end API interface services by including fastAPI and uvicorn components to connect with upstream and downstream data flows;
[0022] Step S1.3.4: Use the cloud-based ERNIE-Bot 4.0 large model to perform multiple rounds of processing on the extracted information, and output the returned results based on the key field information that meets the preset requirements.
[0023] Preferably, the method also includes: the risk information analysis report links the IT asset management CMDB and the host security HIDS, automatically conducts investigations and confirmations in the core system of the financial intranet based on information risk warnings, and explores security threats to internal operating systems, applications, and middleware.
[0024] Preferably, the step S3 adopts:
[0025] Step S3.1: Obtain full flow information based on the network and security communication logs of the financial intranet core system and full flow mirroring;
[0026] Step S3.2: Convert the acquired full traffic information into traffic behavior detection metadata, set the security fields including date and time, source IP, destination IP, access protocol, access port, access traffic, access application, attack detection and security level, and then format and store them uniformly; and divide them into training set and test set;
[0027] Step S3.3: using the training set to train the internal abnormal access behavior detection model;
[0028] Step S3.4: Evaluate the trained internal abnormal access behavior detection model using traditional evaluation indicators, ROC curves, and K-fold cross validation methods;
[0029] Step S3.5: Verify and optimize the parameters of the trained internal abnormal access behavior detection model through the test set;
[0030] Step S3.6: Regularly use real-time data for sampling verification, periodically tune the model configuration parameters according to the actual return situation, and control false positives and missed negatives by adjusting parameters and thresholds.
[0031] Preferably, the method also includes: the trained internal abnormal access behavior detection model detects and judges the attack behavior according to the real-time traffic characteristics, and triggers an alarm when the attack behavior meets the preset conditions; and synchronously sends the alarm information to the corresponding administrator end.
[0032] According to the present invention, a financial core system network security internal threat detection system based on AI includes:
[0033] Module M1: Real-time risk information data uses the trained risk information key information extraction model to extract key information that meets the preset requirements;
[0034] Module M2: The information reporting model generates a risk information analysis report based on the extracted key information that meets the preset requirements;
[0035] Module M3: Real-time full traffic information uses the trained internal abnormal access behavior detection model to detect and judge attack behaviors;
[0036] The risk information key information extraction model is an AI training model built based on the AI integration framework of langchain, the ERNIE-Bot 4.0 large model, and the web framework, with key fields that meet the preset requirements;
[0037] The information reporting model outputs a risk information report based on key information fields through an AI model;
[0038] The internal abnormal access behavior detection model is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network built through a PyTorch framework.
[0039] Preferably, the module M1 adopts:
[0040] Module M1.1: Obtain historical risk information, and pre-process the acquired historical risk information to obtain pre-processed historical risk information;
[0041] Module M1.2: Divide the preprocessed historical information into training set and test set;
[0042] Module M1.3: Use the training set to train the risk information key information extraction model to obtain the trained risk information key information extraction model; use the test set to verify and optimize the parameters of the trained risk key information extraction model;
[0043] Module M1.4: Use test set data and real-time information data to confirm that the trained risk information key information extraction model meets the expectations of risk information key information extraction;
[0044] Module M1.5: Use the trained risk information key information extraction model to extract key field information that meets the preset requirements based on the acquired real-time information data;
[0045] The module M1.1 includes: based on NIST-NVD, CNVD, and security vendor official website announcement information data that meets preset requirements, including cleaning and conversion processing to obtain processed historical risk information data;
[0046] The module M1.3 adopts:
[0047] Module M1.3.1: Compile a document parsing model using components including Pandas, libreoffice, win32com, and PyMuPDF to extract text information from input documents in word, excel, and PDF formats;
[0048] Module M1.3.2: Compile an image recognition model using components including paddlepaddle, opencv, and paddleocr to perform image recognition and text information extraction on input images in jpg format;
[0049] Module M1.3.3: Build front-end and back-end API interface services by including fastAPI and uvicorn components to connect with upstream and downstream data flows;
[0050] Module M1.3.4: Use the cloud-based ERNIE-Bot 4.0 large model to perform multiple rounds of processing on the extracted information, and output the returned results based on the key field information that meets the preset requirements.
[0051] Preferably, the module M3 adopts:
[0052] Module M3.1: Obtain full flow information based on the network and security communication logs of the financial intranet core system and full flow mirroring;
[0053] Module M3.2: Convert the acquired full traffic information into traffic behavior detection metadata, set the security fields including date and time, source IP, destination IP, access protocol, access port, access traffic, access application, attack detection and security level, and then format and store them uniformly; and divide them into training set and test set;
[0054] Module M3.3: Use the training set to train the internal abnormal access behavior detection model;
[0055] Module M3.4: Evaluate the trained internal abnormal access behavior detection model using traditional evaluation indicators, ROC curves, and K-fold cross-validation methods;
[0056] Module M3.5: Verify and optimize the parameters of the trained internal abnormal access behavior detection model through the test set;
[0057] Module M3.6: Regularly use real-time data for sampling verification, periodically tune model configuration parameters based on actual return conditions, and control false positives and missed negatives through parameter and threshold adjustments.
[0058] Compared with the prior art, the present invention has the following beneficial effects:
[0059] 1. The present invention uses an AI big model to conduct unified entry sorting and analysis of network security risk information, solves problems such as complex information source file format, irregular content, unclear files, etc., and outputs the results as input for subsequent security operations, realizing the technology of automatically analyzing risk information and automatically mining internal vulnerabilities;
[0060] 2. The present invention achieves the technical effect of automatically detecting, analyzing and handling internal attack behaviors through the combination of AI deep learning models and multi-model evaluation and optimization methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:
[0062] Figure 1 This is a flowchart of AI model training and optimization according to an embodiment of the present invention;
[0063] Figure 2 This is a flowchart of the actual AI automatic processing described in an embodiment of the present invention. DETAILED DESCRIPTION
[0064] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.
[0065] Example 1
[0066] The purpose of this invention is to overcome the shortcomings of the existing technology in covering scenarios. For the intranet security threats faced by the core system of the financial intranet, through risk information, asset linkage, behavior detection and other modules based on AI construction, linking IT asset management CMDB, operation and maintenance automation disposal module, combined with automatic alarm script tools for real-time reporting, timely and automatically obtain key risk vulnerabilities, self-check internal components, and block internal abnormal behaviors. The final set of efficient and complete network security internal threat detection methods can improve the recognition accuracy and disposal efficiency of internal threats to the network security of the financial core system, and save the execution costs of security operation and maintenance personnel and related IT personnel.
[0067] According to the present invention, a financial core system network security internal threat detection method based on AI includes: Figure 1 As shown, the following steps are included:
[0068] Risk Information Key Information Extraction Model
[0069] S1. Data collection: According to the announcements of NIST-NVD, CNVD, and the official websites of mainstream security vendors, a custom program is compiled to automatically obtain historical risk information, and cooperate with mainstream security vendors to obtain historical data of risk warning push files.
[0070] S2. Initialization of training set: The historical information obtained in step S1 is initially formatted into document files and stored uniformly, and divided into training set and test set.
[0071] S3. Model training: Set key fields such as "date and time", "vulnerability number", "vulnerability level", "vulnerability name", "affected products / components", etc., build an AI training model through the langchain-based AI integration framework, ERNIE-Bot 4.0 large model, and common web framework, and use the training set data for AI model training.
[0072] S3.1 Document parsing: A document parsing program is compiled through Pandas, libreoffice, win32com, and PyMuPDF components to extract text information from input documents in word, excel, PDF and other formats.
[0073] S3.2 Image recognition: Use paddlepaddle, opencv, and paddleocr components to compile an image recognition program to perform image recognition and extract text information from input images in jpg and other formats.
[0074] S3.3 Web framework: Build front-end and back-end API interface services through fastAPI and uvicorn components to connect with upstream and downstream data streams.
[0075] S3.4 Large model: Use the cloud-based ERNIE-Bot 4.0 large model to perform multiple rounds of processing on the extracted information and output the returned results based on the key field parameters.
[0076] S4. Model tuning: Use the test set data for validation, configuration, and parameter tuning.
[0077] S5. Return verification: Use test set data and formal environment data to confirm that the AI model meets the expectations for extracting key information from risk information. At this point, the training of the risk information key information extraction model is completed.
[0078] Internal abnormal access behavior detection model
[0079] S6.: Based on the network and security communication logs of the financial intranet core system and the full-flow mirror, compile an automated program to obtain full-flow information.
[0080] S7. Initialize the training set: Format the historical traffic syslog data obtained in step S6 into traffic behavior detection metadata, set security fields such as "date and time", "source IP", "destination IP", "access protocol", "access port", "access traffic", "access application", "attack detection", and "security level" and then format and store them uniformly. After manual intranet attack behavior identification, they are divided into training set and test set.
[0081] S8. Model training: Convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory network (LSTM) and other models are built through the PyTorch framework to perform model training on the training set.
[0082] S9. Model tuning: Verify the model algorithm by combining traditional evaluation indicators, ROC curve, and K-fold cross validation method:
[0083] S9.1 Calculation of traditional evaluation indicators: The actual number of positive samples predicted by the statistical model is tp; the actual number of negative samples predicted by the statistical model is fp; the actual number of negative samples predicted by the statistical model is tn; the actual number of positive samples predicted by the statistical model is fn. Evaluation score:
[0084] SA1=50%*Acc+50%*FS=50%*(tp+tn) / (tp+fn+tn+fp)+50%*2*[tp*tp / (tp+fp)*(fp+fn)] / [tp / (tp+fp)+tp / (tp+fn)], calculate SA1_CNN, SA1_RNN, SA1_LSTM respectively, and compare the traditional indicators as the most intuitive results, with an overall weight of 60%. SC1_CNN=SA1_CNN / MAX(SA1),SC1_RNN=SA1_RNN / MAX(SA1),SC1_LSTM=SA1_LSTM / MAX(SA1).
[0085] S9.2 ROC curve calculation: Calculate FPR = fp / fp+tn, TPR = tp / tp+fn respectively, call pythonsklearn package to calculate AUC, get evaluation index SA2 = metrics.auc(FPR, TPR), calculate SA2_CNN, SA2_RNN, SA2_LSTM respectively, with an overall weight of 40%. SC2_CNN = SA2_CNN / MAX(SA2), SC2_RNN = SA2_RNN / MAX(SA2), SC2_LSTM = SA2_LSTM / MAX(SA2).
[0086] S9.3 K-fold cross validation calculation and evaluation summary: The samples are divided and cross-validated using a 10-fold cross validation method, and the arithmetic average is taken and weighted to obtain the comprehensive evaluation result:
[0087] SC_CNN=60%*AVE(SC1_CNN(1,2,3…10))+40%*AVE(SC2_CNN(1,2,3…10))
[0088] SC_RNN=60%*AVE(SC1_RNN(1,2,3…10))+40%*AVE(SC2_RNN(1,2,3…10))
[0089] SC_LSTM=60%*AVE(SC1_LSTM(1,2,3…10))+40%*AVE(SC2_LSTM(1,2,3…10))
[0090] The selection and configuration of the classification model and threshold tuning are confirmed based on the total score of the verification effect of the test set, and the optimal model TPR is made greater than 90%, and the confidence threshold is set to 90%.
[0091] S10. Return verification: Use test set data and formal environment data to confirm that the AI model meets the expectations of intranet threat behavior detection. At this point, the training of the internal abnormal access behavior detection model is completed.
[0092] Regular optimization scripts for operation and maintenance
[0093] S11. Regular model optimization: Regularly use real-time data for sampling verification, periodically tune model configuration parameters based on actual return conditions, and control false positives and missed positives by adjusting parameters and thresholds.
[0094] See also Figure 2 , the actual AI automatic processing process includes the following steps:
[0095] S12. Real-time information data collection: According to the announcements of NIST-NVD, CNVD, and the official websites of mainstream security vendors, we compile custom programs to automatically obtain real-time risk information, and cooperate with mainstream security vendors to obtain real-time push files of risk warnings.
[0096] S13. Model processing: Real-time data is automatically transmitted to the trained risk information key information extraction model.
[0097] S14. Result output: The AI model returns key field information based on real-time risk information.
[0098] S15. Information report: The information report model regularly issues complete and accurate risk information analysis reports based on the key field information of the information returned by the AI model, helping administrators to obtain formatted hot security risk information from multiple sources in a timely manner.
[0099] S16. Component management: Information reports are linked to IT asset management CMDB and host security HIDS systems, automatically based on information risk warnings, to conduct checks and confirmations in the core systems of the financial intranet, and to uncover security threats to internal operating systems, applications, and middleware.
[0100] S17. Real-time traffic data collection: Based on the network and security communication logs of the financial intranet core system and the full traffic mirror, an automated program is compiled to obtain real-time full traffic information.
[0101] S18. Model processing: Real-time data is automatically transmitted to the trained internal abnormal access behavior detection model.
[0102] S19. Result output: The AI model detects and judges attack behaviors based on real-time traffic characteristics, and triggers alarms and takes measures for traffic behaviors that meet the threshold conditions, that is, those with a probability of being an attack >90%.
[0103] S20. Information synchronization: Push internal security threat information from S15, S16, and S19 to corresponding security operation and maintenance personnel, system administrators, and network administrators through notification methods such as email, WeChat, and Lark.
[0104] S21. Exception handling:
[0105] 1) Conduct internal team announcements and risk warnings for risk information that meets the requirements.
[0106] 2) For host or system vulnerabilities that meet the conditions, the host security vulnerability repair function and vulnerability virtual patch function are triggered to automatically deal with the threats.
[0107] 3) For internal attack behaviors that meet the conditions, the firewall and other modules are automatically linked to take emergency measures through temporary or long-term blacklists.
[0108] The present invention also provides an AI-based financial core system network security internal threat detection system. The AI-based financial core system network security internal threat detection system can be implemented by executing the process steps of the AI-based financial core system network security internal threat detection method, that is, those skilled in the art can understand the AI-based financial core system network security internal threat detection method as a preferred implementation of the AI-based financial core system network security internal threat detection system.
[0109] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.
[0110] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.
Claims
1. An AI-based method for detecting internal threats in the network security of financial core systems. It is characterized in that include: Step S1: Real-time risk information data uses the trained risk information key information extraction model to extract key information that meets preset requirements; Step S2: The information reporting model generates a risk information analysis report based on the extracted key information that meets the preset requirements; Step S3: Real-time full traffic information uses the trained internal abnormal access behavior detection model to detect and judge attack behaviors; The risk information key information extraction model is an AI training model built based on the AI integration framework of langchain, the ERNIE-Bot 4.0 large model, and the web framework, with key fields that meet the preset requirements; The information reporting model outputs a risk information report based on key information fields through an AI model; The internal abnormal access behavior detection model is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network built through a PyTorch framework.
2. According to claim 1, the AI-based financial core system network security internal threat detection method, It is characterized in that The step S1 adopts: Step S1.1: Acquire historical risk information, and preprocess the acquired historical risk information to obtain preprocessed historical risk information; Step S1.2: Divide the preprocessed historical information into a training set and a test set; Step S1.3: Use the training set to train the risk information key information extraction model to obtain the trained risk information key information extraction model; use the test set to verify and optimize the parameters of the trained risk key information extraction model; Step S1.4: Use the test set data and real-time information data to confirm that the trained risk information key information extraction model meets the risk information key information extraction expectations; Step S1.5: Utilize the trained risk information key information extraction model to extract key field information that meets preset requirements based on the acquired real-time information data.
3. The AI-based financial core system network security internal threat detection method according to claim 2, It is characterized in that The step S1.1 includes: performing cleaning and conversion processing based on NIST-NVD, CNVD, and security vendor official website announcement information data that meets preset requirements to obtain processed historical risk information data.
4. The AI-based financial core system network security internal threat detection method according to claim 2, It is characterized in that The step S1.3 adopts: Step S1.3.1: compile a document parsing model using components including Pandas, libreoffice, win32com, and PyMuPDF to extract text information from input documents in word, excel, and PDF formats; Step S1.3.2: compile an image recognition model using components including paddlepaddle, opencv, and paddleocr to perform image recognition and text information extraction on input images in jpg format; Step S1.3.3: Build front-end and back-end API interface services by including fastAPI and uvicorn components to connect with upstream and downstream data flows; Step S1.3.4: Use the cloud-based ERNIE-Bot 4.0 large model to perform multiple rounds of processing on the extracted information, and output the returned results based on the key field information that meets the preset requirements.
5. The AI-based financial core system network security internal threat detection method according to claim 1, It is characterized in that The method also includes: risk information analysis reports linking IT asset management CMDB and host security HIDS, automatically conducting investigations and confirmations in the core systems of the financial intranet based on information risk warnings, and exploring security threats to internal operating systems, applications, and middleware.
6. The AI-based financial core system network security internal threat detection method according to claim 1, It is characterized in that The step S3 adopts: Step S3.1: Obtain full flow information based on the network and security communication logs of the financial intranet core system and full flow mirroring; Step S3.2: Convert the acquired full traffic information into traffic behavior detection metadata, set the security fields including date and time, source IP, destination IP, access protocol, access port, access traffic, access application, attack detection and security level, and then format and store them uniformly; And divided into training set and test set; Step S3.3: using the training set to train the internal abnormal access behavior detection model; Step S3.4: Evaluate the trained internal abnormal access behavior detection model using traditional evaluation indicators, ROC curves, and K-fold cross validation methods; Step S3.5: Verify and optimize the parameters of the trained internal abnormal access behavior detection model through the test set; Step S3.6: Regularly use real-time data for sampling verification, periodically tune the model configuration parameters according to the actual return situation, and control false positives and missed negatives by adjusting parameters and thresholds.
7. The AI-based financial core system network security internal threat detection method according to claim 1, It is characterized in that The method also includes: the trained internal abnormal access behavior detection model detects and judges the attack behavior according to the real-time traffic characteristics, and when the attack behavior meets the preset conditions, an alarm is triggered; and the alarm information is synchronously sent to the corresponding administrator end.
8. An AI-based internal threat detection system for financial core system network security, It is characterized in that include: Module M1: Real-time risk information data uses the trained risk information key information extraction model to extract key information that meets the preset requirements; Module M2: The information reporting model generates a risk information analysis report based on the extracted key information that meets the preset requirements; Module M3: Real-time full traffic information uses the trained internal abnormal access behavior detection model to detect and judge attack behaviors; The risk information key information extraction model is an AI training model built based on the AI integration framework of langchain, the ERNIE-Bot 4.0 large model, and the web framework, with key fields that meet the preset requirements; The information reporting model outputs a risk information report based on key information fields through an AI model; The internal abnormal access behavior detection model is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network built through a PyTorch framework.
9. The AI-based financial core system network security internal threat detection system according to claim 8, It is characterized in that The module M1 adopts: Module M1.1: Obtain historical risk information, and pre-process the acquired historical risk information to obtain pre-processed historical risk information; Module M1.2: Divide the preprocessed historical information into training set and test set; Module M1.3: Use the training set to train the risk information key information extraction model to obtain the trained risk information key information extraction model; use the test set to verify and optimize the parameters of the trained risk key information extraction model; Module M1.4: Use test set data and real-time information data to confirm that the trained risk information key information extraction model meets the expectations of risk information key information extraction; Module M1.5: Use the trained risk information key information extraction model to extract key field information that meets the preset requirements based on the acquired real-time information data; The module M1.1 includes: based on NIST-NVD, CNVD, and security vendor official website announcement information data that meets preset requirements, including cleaning and conversion processing to obtain processed historical risk information data; The module M1.3 adopts: Module M1.3.1: Compile a document parsing model using components including Pandas, libreoffice, win32com, and PyMuPDF to extract text information from input documents in word, excel, and PDF formats; Module M1.3.2: Compile an image recognition model using components including paddlepaddle, opencv, and paddleocr to perform image recognition and text information extraction on input images in jpg format; Module M1.3.3: Build front-end and back-end API interface services by including fastAPI and uvicorn components to connect with upstream and downstream data flows; Module M1.3.4: Use the cloud-based ERNIE-Bot 4.0 large model to perform multiple rounds of processing on the extracted information, and output the returned results based on the key field information that meets the preset requirements.
10. The AI-based financial core system network security internal threat detection system according to claim 8, It is characterized in that The module M3 adopts: Module M3.1: Obtain full flow information based on the network and security communication logs of the financial intranet core system and full flow mirroring; Module M3.2: Convert the acquired full traffic information into traffic behavior detection metadata, set the security fields including date and time, source IP, destination IP, access protocol, access port, access traffic, access application, attack detection and security level, and then format and store them uniformly; And divided into training set and test set; Module M3.3: Use the training set to train the internal abnormal access behavior detection model; Module M3.4: Evaluate the trained internal abnormal access behavior detection model using traditional evaluation indicators, ROC curves, and K-fold cross-validation methods; Module M3.5: Verify and optimize the parameters of the trained internal abnormal access behavior detection model through the test set; Module M3.6: Regularly use real-time data for sampling verification, periodically tune model configuration parameters based on actual return conditions, and control false positives and missed negatives through parameter and threshold adjustments.
